Abstract 5358: Digital SP263 PD-L1 tumor cell scoring in non-small cell lung cancer achieves comparable outcome prediction to manual pathology scoring

Abstract

Abstract Background: Tumor cell (TC) PD-L1 expression is predictive of response to PD-L1-targeted immunotherapy, and accurate scoring is crucial for treatment selection. Scoring relies on manual assessment of immunohistochemically labeled tissue and is subject to subjective variation due to pathologist assessment. As a digital alternative, a clone-agnostic AI-based model for PD-L1 quantification in non-small cell lung cancer (AIM-PD-L1 NSCLC) was developed1. AIM-PD-L1 was deployed on samples from a Phase 3 study of anti-PD-L1 atezolizumab combination therapy with carboplatin and paclitaxel, and/or bevacizumab in Stage IV NSCLC (IMpower150; NCT02366143). Digital and manual PD-L1 TC scores were compared and interrogated for their respective potential to predict efficacy to atezolizumab combination treatments. Methods: AIM-PD-L1 was deployed on images (n=768) digitized from SP263-labeled slides with available manual pathologist TC scores to quantify tissue regions and individual TCs. PD-L1 expression status was determined for each tumor cell and a slide-level TC score was computed. Overall survival (OS) and progression free survival (PFS) analyses of patients at selected cutoffs of 1%, 50%, and across a continuum of cutoffs from 0% to 100% PD-L1 TC by digital and manual methods were conducted, comparing groups treated with or without atezolizumab in combinations with carboplatin and paclitaxel, and/or bevacizumab. Digital and manual scores were compared using agreement rates, Lin's concordance and Spearman’s correlation coefficients, and hazard ratios (HRs) were calculated for OS and PFS analysis. Results: At the slide level, correlation between continuous digital and manual scores was high (r 0.84 [95% CI 0.81-0.86]). Digital assessment of PD-L1 positivity at the 1% cutoff identified more positive patients than manual scoring (70% vs. 55% prevalence, respectively), with comparable treatment benefit (digital OS HR 0.69 [0.53-0.9], and manual OS HR 0.72 [0.54-0.98], digital PFS HR 0.67 [0.54-0.83], and manual PFS HR 0.66 [0.52-0.84]). Continuous digital PD-L1 TC scores showed that treatment benefit improved for patients with scores ≥50% compared to manual scoring. At the ≥50% PD-L1 TC cutoff, numerical improvement was observed in OS and PFS by digital scoring compared to manual (digital OS HR 0.5 [0.29-0.86] compared to manual OS HR 0.64 [0.4-1.02], and digital PFS HR 0.37 [0.24-0.57], compared to manual PFS HR 0.53 [0.37-0.76]). Conclusions: AIM-PD-L1 scoring was as effective at predicting outcomes as manual using the ≥1% PD-L1 TC expression and at the ≥50% TC cutoff, digital scoring identified a subgroup with enriched efficacy compared to manual. Further evaluation of the accuracy and reproducibility of PD-L1 scoring by digital pathology as well as its potential use for patient enrollment or stratifications in clinical trials is needed. 1Griffin et. al. AACR (2022) Citation Format: Hen Prizant, John Shamshoian, John Abel, Andrew Beck, Laura Chambre, Stephanie Hennek, Hartmut Koeppen, Daniel Ruderman, Meghna Das Thakur, Michael Montalto, Benjamin Trotter, Ilan Wapinski, Wei Zou, Minu K. Srivastava, Jennifer Giltnane. Digital SP263 PD-L1 tumor cell scoring in non-small cell lung cancer achieves comparable outcome prediction to manual pathology scoring. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5358.